Global Spatial Autocorrelation
Global Spatial Autocorrelation Analysis · Also known as: global spatial dependence, global Moran's I, GSA, global spatial clustering measure
Global Spatial Autocorrelation measures the degree to which similar values cluster together across an entire study area. Rather than identifying where clusters occur, it yields a single summary statistic — most commonly Moran's I — that quantifies whether spatial proximity coincides with value similarity, dissimilarity, or randomness across all observations simultaneously.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use Global Spatial Autocorrelation as a first-pass diagnostic before any spatial modeling: it confirms whether your data exhibit spatial dependence that must be accounted for. It is appropriate for continuous or count attributes measured at areal units (census tracts, counties) or interpolated point supports when n is at least 30 and a meaningful spatial weights matrix can be defined. Do not rely solely on the global statistic when you suspect heterogeneous clustering patterns — local statistics are needed then. It is also not suitable when the study area boundary is arbitrary and cuts through natural clusters (edge effects), or when observations are not spatially referenced.
Strengths & limitations
- Provides a single, interpretable summary of spatial dependence for the entire study area.
- Moran's I has a well-understood expected value and variance under spatial randomness, enabling formal hypothesis testing.
- Permutation-based inference is robust to non-normal attribute distributions.
- Computationally efficient even for large datasets when using sparse weights matrices.
- Serves as an essential diagnostic step before fitting spatial regression models.
- Returns only a single global summary — local clusters of opposite sign can cancel out, masking heterogeneous spatial patterns.
- Results are sensitive to the choice of spatial weights matrix; different neighborhood definitions can yield different conclusions.
- Assumes stationarity — that spatial dependence is uniform across the study area, which is often unrealistic.
- Does not identify where clusters are located; follow-up local analysis is always required for that purpose.
- Modified Moran's I variants are needed for non-Gaussian outcomes (binary, count, or rate data).
Frequently asked
What is the range of Moran's I and how do I interpret its sign?
Moran's I ranges approximately from -1 to +1. Positive values near +1 indicate strong positive spatial autocorrelation (similar values cluster together). Values near -1 indicate negative autocorrelation (dissimilar values are neighbors). Values near 0 (or the expected value -1/(n-1)) suggest spatial randomness.
How does Global Spatial Autocorrelation differ from Local Moran's I?
The global statistic produces a single value summarizing spatial dependence across the whole study area. Local Moran's I (a LISA statistic) decomposes this into location-specific values, revealing where significant clusters or outliers occur — high-high clusters, low-low clusters, and spatial outliers.
Which spatial weights matrix should I use?
The choice should be theoretically motivated. Contiguity weights (queen or rook) suit areal polygon data. K-nearest-neighbor or distance-band weights suit point data or irregular polygon shapes. Sensitivity analysis across two or three plausible specifications is good practice.
Should I use the normal approximation or permutation inference?
Permutation inference (typically 999 random permutations) is generally preferred because it makes no distributional assumption about the attribute. The normal approximation is faster but can be misleading for skewed or heavy-tailed distributions.
A significant Moran's I — do I now need a spatial regression model?
A significant global Moran's I in your OLS residuals (not just your raw variable) indicates that the spatial dependence is not fully explained by your covariates and that ignoring it will bias inference. In that case, switching to a spatial lag or spatial error model is advisable.
Sources
- Moran, P. A. P. (1950). Notes on continuous stochastic phenomena. Biometrika, 37(1/2), 17–23. DOI: 10.2307/2332142 ↗
- Anselin, L. (1988). Spatial Econometrics: Methods and Models. Kluwer Academic Publishers. ISBN: 978-9024737322
How to cite this page
ScholarGate. (2026, June 3). Global Spatial Autocorrelation Analysis. ScholarGate. https://scholargate.app/en/spatial-analysis/global-spatial-autocorrelation
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Geary's CSpatial analysis↔ compare
- Hot Spot AnalysisSpatial analysis↔ compare
- Local Moran's ISpatial analysis↔ compare
- Local Spatial AutocorrelationSpatial analysis↔ compare
- Moran's ISpatial analysis↔ compare
- Spatial AutocorrelationSpatial analysis↔ compare